Talk

FuzzyAI: Attacking LLMs with Coverage-Guided Fuzzing

April 25, 10:30 (CAMPUS)

With Large Language Models (LLMs) like ChatGPT, Bard, and Claude swiftly establishing themselves as keystones in our digital ecosystem, the inevitable is on the horizon: an explosion of adversarial attacks targeting these systems, leading to severe data leaks and misguided outputs. Leveraging our profound experience in vulnerability research and a robust background in the bug bounty community, our team has pivoted to address the nuances of LLMs. Our intent doesn’t halt at mere identification; we’re pioneering the generation of these potential adversarial attacks. Central to our strategy is the amalgamation of GaN-based fuzzers and attention-centric detection tools. In this session, attendees will be offered an immersive journey, marrying traditional vulnerability research techniques with the evolving demands of LLM security, thereby sketching a roadmap for the future of adversarial defense strategies.

Speaker

Eran Shimony

Eran Shimony is a Principal Security Researcher at CyberArk with an extensive background in security research that includes years of experience in malware analysis and vulnerability research on multiple platforms. He previously spoke at RSAC, Nullcon, HITB Amsterdam, and many more. Shimony has discovered several dozen acknowledged vulnerabilities across major vendors including Microsoft, Intel, Samsung, Facebook, and many more. Besides finding security bugs, he enjoys mixing and, of course, drinking cocktails.

Mark Cherp

Mark Cherp is a Vulnerability Team Leader at CyberArk with a special interest in AI and low-level, kernel-space attack vectors and a strong interest in fuzzing and other automation techniques for bug discovery. Mark has previously worked for Microsoft, Checkpoint, and several other companies in the Israeli cyber industry. He had the chance to tackle multiple vulnerability research domains such as cloud, network, mobile, and other endpoints.

Organized by

Technology partners

Partner events

Scroll to Top